# Microsoft Foundry Managed Compute reaches general availability for open models

published: 2026-11-20T19:05:50.218Z
topic: infrastructure
source: https://devblogs.microsoft.com/foundry/managed-compute-ga/
url: https://aiagentnews.io/stories/microsoft-foundry-managed-compute-reaches-general-availability-for-open-models

Managed Compute in Microsoft Foundry is now generally available, following its preview at Microsoft Build in June. The service provisions GPUs, maintains the serving containers and operates the runtime so teams can serve open models on dedicated hardware.

Microsoft has made Managed Compute in Microsoft Foundry generally available after previewing it in June. Users pick one of more than 150 open models from the catalog, then a deployment template that sets the context length, serving runtime and GPU configuration. Depending on the model, deployments run on NVIDIA A100, H100 or H200 GPUs, or on AMD MI300X GPUs.

Capacity is added by increasing the instance count behind an existing endpoint, and each instance uses the GPUs its template specifies. In one test, a five-instance DeepSeek V4.1 Flash deployment processed about five times as many tokens as a single instance and returned fewer HTTP 429 responses. Foundry also offers prompt-aware and load-aware routing, and a Signal65 benchmark compared it with a tuned self-hosted setup on MI300X virtual machines.

Custom LoRA adapter deployment is in preview, and bring-your-own SafeTensors weights are in limited public preview. Billing is hourly for each accelerator.

## Why it matters if you build agents

Developers can run open models on dedicated GPUs behind the same Foundry endpoints they already call, without managing containers or serving runtimes themselves. Costs scale with instance count, since each instance multiplies GPU usage and hourly charges. Check GPU quota before deploying, and measure token throughput, status codes and streaming latency on your own workload before sizing a deployment.
